An electronic apparatus performs a method of updating an inter-predicted current block using a neighboring affine block. The electronic apparatus first identifies a pixel within the inter-predicted current block, the pixel having a first inter-predicted pixel value. Next, the electronic apparatus determines a motion vector difference for the pixel based on a set of affine parameters of the neighboring affine block and then a pixel value difference for the pixel according to the motion vector difference. The pixel value difference is an inner product of the pixel value gradient vector and the motion vector difference as the pixel value difference. Finally, the electronic apparatus updates the first inter-predicted pixel value with the pixel value difference as a second inter-predicted pixel value.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
3. The method of claim 2, wherein the pixel value gradient vector for the pixel is estimated by applying a 3-tap filter to pixel values at the pixel and its neighboring pixels within the inter-predicted current block.
This invention relates to video encoding, specifically improving motion compensation by refining pixel value gradients within inter-predicted blocks. The problem addressed is the need for more accurate gradient estimation to enhance motion vector prediction and reduce encoding artifacts. The method involves estimating a pixel value gradient vector for a pixel within an inter-predicted block by applying a 3-tap filter to the pixel values of the pixel and its neighboring pixels. The 3-tap filter processes these values to compute a gradient vector that represents the directional change in pixel intensity. This gradient vector is then used to refine motion vector prediction, improving the accuracy of motion compensation and reducing visual distortions in the encoded video. The neighboring pixels considered for gradient estimation are those within the same inter-predicted block, ensuring local consistency. The method is particularly useful in high-efficiency video coding (HEVC) and similar standards where precise motion estimation is critical for compression efficiency. By leveraging local pixel gradients, the technique enhances the prediction process without requiring additional computational overhead, making it suitable for real-time encoding applications.
4. The method of claim 1, wherein the sub-block is located at an upper section of the inter-predicted current block and the neighboring affine block is located above the inter-predicted current block.
This invention relates to video encoding and decoding, specifically improving motion compensation for inter-predicted blocks in affine motion models. The problem addressed is inefficient motion compensation when an inter-predicted block shares a boundary with an affine-predicted block, leading to inaccuracies in motion estimation and reconstruction. The method involves refining motion compensation for an inter-predicted current block by considering a neighboring affine block. The inter-predicted current block is divided into sub-blocks, and at least one sub-block is positioned in the upper section of the current block. The neighboring affine block is located directly above the inter-predicted current block. Motion vectors or affine parameters from the neighboring affine block are used to refine the motion compensation of the upper sub-block of the current block. This ensures smoother transitions and more accurate motion representation at block boundaries, improving compression efficiency and visual quality. The technique leverages spatial correlation between adjacent blocks, particularly when one block uses affine prediction while the other uses traditional inter prediction. By incorporating information from the affine block, the method enhances the accuracy of motion compensation for the inter-predicted block, reducing artifacts and improving coding performance. This approach is particularly useful in video sequences with complex motion patterns where affine models provide better motion representation.
5. The method of claim 1, wherein the sub-block is located at a left section of the inter-predicted current block and the neighboring affine block is located left to the inter-predicted current block.
This invention relates to video encoding and decoding, specifically improving motion compensation for inter-predicted blocks in video compression. The problem addressed is inefficient motion compensation when predicting blocks using affine motion models, particularly when neighboring blocks have different motion characteristics. The method involves refining motion compensation for an inter-predicted current block by using information from a neighboring affine block. The current block is divided into sub-blocks, and at least one sub-block is positioned in the left section of the current block. The neighboring affine block is located immediately to the left of the current block. The affine block provides motion information that helps predict the motion of the left sub-block of the current block more accurately. This approach improves prediction efficiency by leveraging spatial correlation between adjacent blocks, reducing residual errors and enhancing compression performance. The method is particularly useful in video coding standards that support affine motion models, such as HEVC and its extensions. By refining motion compensation in this manner, the invention reduces bitrate while maintaining or improving video quality.
6. The method of claim 1, wherein the inter-predicted current block and the neighboring affine block have the same reference picture.
This invention relates to video encoding and decoding, specifically improving motion compensation in affine motion prediction. The problem addressed is inefficient motion compensation when predicting a current block using affine motion models, particularly when neighboring blocks share the same reference picture but are not properly utilized for prediction. The method involves predicting a current block using affine motion compensation, where the current block is an inter-predicted block and a neighboring block is an affine block. The key improvement is ensuring that the current block and the neighboring affine block reference the same picture, allowing for more accurate motion prediction. This avoids inconsistencies that arise when different reference pictures are used, leading to better compression efficiency and reduced bitrate. The neighboring affine block provides motion information that is used to predict the current block's motion parameters. By enforcing the same reference picture for both blocks, the method ensures that the motion vectors and affine parameters are derived from a consistent reference, improving prediction accuracy. This is particularly useful in scenarios where affine motion models are applied, such as in video sequences with complex motion patterns like rotation or zooming. The technique enhances existing affine motion prediction by leveraging spatial correlations between blocks, reducing redundancy and improving compression performance. It is applicable in video coding standards like HEVC or VVC, where affine motion compensation is used to handle non-translational motion.
8. The method of claim 7, wherein the weighting factor is less than 1.
This invention relates to a method for optimizing data processing in a system where multiple data sources contribute to a final output. The problem addressed is the need to balance the influence of different data sources to improve accuracy and reliability of the final result. The method involves assigning a weighting factor to each data source, where the weighting factor determines the relative importance of that source in the final output. The weighting factor is set to be less than 1, meaning no single data source dominates the result, ensuring a more balanced and robust outcome. The method also includes dynamically adjusting the weighting factors based on the quality or reliability of each data source, allowing the system to adapt to changing conditions. This dynamic adjustment helps maintain optimal performance even when some data sources become less reliable or more reliable over time. The method is particularly useful in applications where multiple sensors, algorithms, or data streams contribute to a decision-making process, such as in autonomous systems, predictive analytics, or sensor fusion applications. By ensuring that no single source has an overwhelming influence, the method improves the overall reliability and accuracy of the system.
9. The method of claim 1, wherein the neighboring affine block is a first one of a plurality of neighboring blocks of the inter-predicted current block having an affine mode according to a predefined order.
This invention relates to video encoding and decoding, specifically improving affine motion compensation for inter-predicted blocks in video compression. The problem addressed is efficiently determining neighboring affine blocks to enhance prediction accuracy while reducing computational complexity. The method involves selecting a neighboring affine block from multiple candidate blocks surrounding an inter-predicted current block. The selection follows a predefined order to ensure consistent and efficient processing. The neighboring affine block must itself be encoded or decoded using an affine mode, which allows for more flexible motion modeling compared to traditional block-based methods. By leveraging the motion information from this neighboring affine block, the current block's motion can be more accurately predicted, improving compression efficiency. The predefined order ensures that the selection process is deterministic and reproducible, which is critical for synchronization between encoder and decoder. This approach helps maintain low computational overhead while improving prediction quality, particularly in regions with complex motion patterns. The method is part of a broader system for video coding that includes affine motion compensation techniques to handle non-translational motion more effectively.
10. The method of claim 1, wherein the inter-predicted current block is a bi-predictive block.
This invention relates to video encoding and decoding, specifically improving bi-predictive block processing in inter-frame prediction. The problem addressed is inefficient handling of bi-predictive blocks, which use motion vectors from two reference frames to improve compression but can introduce artifacts or computational overhead. The method involves processing a bi-predictive block, which is a current block in a video frame predicted using motion vectors from two reference frames. The technique optimizes the prediction process by refining how motion vectors are derived or combined for bi-predictive blocks. This may include techniques like weighted averaging of motion vectors, adaptive selection of reference frames, or improved interpolation methods to reduce prediction errors. The goal is to enhance compression efficiency while maintaining or improving video quality. The method may also involve additional steps such as motion vector refinement, where the initial motion vectors are adjusted based on local or global motion analysis. This refinement can help reduce prediction errors, especially in areas with complex motion. The technique may further include adaptive weighting of the two motion vectors to better handle occlusions or overlapping motion. The overall approach aims to balance computational complexity with prediction accuracy, making it suitable for real-time or high-efficiency video coding applications.
11. The method of claim 1, wherein the inter-predicted current block is a uni-predictive block.
A video encoding method improves compression efficiency by refining the prediction process for uni-predictive blocks. In video coding, inter-prediction reduces redundancy by predicting a current block using motion-compensated reference blocks from previously encoded frames. Uni-predictive blocks rely on a single prediction direction (forward or backward) rather than bidirectional prediction, which can be less accurate but reduces computational overhead. The method enhances uni-predictive block encoding by optimizing the prediction process, such as refining motion vector derivation or improving reference frame selection. This may involve adaptive techniques to select the most efficient prediction mode or adjusting prediction parameters based on block characteristics. The goal is to balance compression efficiency and computational complexity, particularly for blocks where bidirectional prediction is unnecessary or less effective. The method ensures that uni-predictive blocks are encoded with minimal residual data while maintaining visual quality, contributing to overall bitrate reduction in video streams.
12. The method of claim 1, wherein at least one pixel value within the inter-predicted current block is not updated by any neighboring affine block.
This invention relates to video encoding and decoding, specifically improving motion compensation in affine motion models. The problem addressed is inefficient motion compensation in video coding, where inter-predicted blocks may not fully utilize neighboring affine blocks for accurate prediction, leading to redundant computations and suboptimal compression. The method involves refining motion compensation for a current block in a video frame by selectively updating pixel values based on neighboring affine blocks. Unlike traditional approaches, this method ensures that at least one pixel value within the inter-predicted current block remains unchanged by any neighboring affine block, preventing unnecessary updates that could degrade prediction accuracy. The neighboring affine blocks are used to derive motion vectors or control points, which are then applied to the current block to improve prediction efficiency. The method may also involve determining whether neighboring blocks are affine-coded and using their motion information to refine the current block's prediction. This selective updating reduces computational overhead while maintaining or improving prediction quality, particularly in regions with complex motion patterns. The approach is applicable in video codecs like HEVC or VVC, where affine motion models are used to enhance compression efficiency.
15. The electronic apparatus of claim 14, wherein the pixel value gradient vector for the pixel is estimated by applying a 3-tap filter to pixel values at the pixel and its neighboring pixels within the inter-predicted current block.
This invention relates to image processing, specifically to techniques for estimating pixel value gradients in inter-predicted video blocks to improve motion compensation and prediction accuracy. The problem addressed is the need for efficient and accurate gradient estimation in video encoding, particularly for blocks where motion prediction is applied. Traditional methods may suffer from inaccuracies or computational inefficiency when estimating gradients for inter-predicted blocks. The invention describes an electronic apparatus configured to process video data, where the apparatus includes a gradient estimation module. This module estimates a pixel value gradient vector for a pixel within an inter-predicted current block by applying a 3-tap filter to the pixel values of the pixel and its neighboring pixels within the same block. The 3-tap filter is a lightweight computational technique that reduces complexity while maintaining gradient accuracy. The neighboring pixels used for gradient estimation are those adjacent to the current pixel within the inter-predicted block, ensuring local consistency in gradient calculations. This approach enhances motion prediction by providing more precise gradient information, which is critical for accurate block matching and residual reduction in video compression. The method is particularly useful in video encoding standards where efficient gradient estimation is required to balance computational cost and prediction performance.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
November 15, 2021
March 26, 2024
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.